Cybercriminals are getting smarter, outpacing traditional fraud detection methods. As a result, the conversation is now shifting towards AI-driven fraud detection in banking.
The Association of Certified Fraud Examiners found that fraud costs organisations an estimated 5% of their revenue, amounting to billions annually. Hence, it is crucial for banks to prevent fraud before it occurs. This can be achieved if financial institutions incorporate AI into their fraud-detection systems.
Speaking at the DECODE webinar, Mayank Gupta, the head of Middle Office Technology at DBS Tech India, said, “Traditionally, fraud detection relied heavily on rule-based systems. Regulators identified and tracked well-known patterns and familiar modus operandi using predefined rules. However, with the advancements in technology, AI, and payment systems, the landscape of financial transactions has evolved.”
Historically, fraud was identified only after transactions were completed. Now, financial institutions are pivoting to a proactive, real-time approach, aiming to detect suspicious activity mid-transaction.
Are Fraudsters Smarter than Banks?
Yet, as banks level up their defences, fraudsters are becoming more adept at bypassing them. Sahil Aneja, vice president and consumer insights head at HDFC Bank, pointed out that traditional rule-based monitoring, though foundational, is rigid and struggles to keep up with the fraudsters’ evolving methods.
“For example, when UPI was launched, there was a significant spike in fraud incidents. Banks responded by setting thresholds for unusual transactions, which temporarily reduced the scale and frequency of such fraud. However, fraudsters soon adjusted to these rules, necessitating a shift to AI-driven platforms for better fraud detection,” he said.
Transitioning to AI has brought machine learning models like Random Forest and XGBoost, which can detect patterns more effectively. However, these models face challenges due to data imbalances as fraudulent transactions vastly outnumber legitimate ones.
Advanced techniques like neural networks, LSTMs, GANs, and generative AI have recently enhanced fraud detection by identifying long-term behavioural patterns in transactions. Attention mechanisms introduced in 2017, for instance, help models track these behaviours, improving fraud prevention.
What is Fraud Detection?
Aneja spoke about a general benchmark in fraud prevention, indicating that within 30-45 days, fraudsters adapt and develop new methods to bypass platforms put in place by financial institutions. This means that institutions must continuously refine their systems, transitioning to self-learning models.
Maintaining up-to-date fraud detection can get challenging as manual recalibration in response to evolving fraud techniques can exhaust teams and strain systems. Thus, significant resources are being invested in building self-learning models that study undetected fraud patterns, recalibrate accordingly, and redeploy the updated models automatically.
Implementing AI in Fraud Detection
Two core elements are crucial for AI-based fraud detection to work effectively: robust data engineering and advanced platform capabilities. Real-time data streaming and timely model updates allow institutions to catch fraud within milliseconds. Building infrastructure that supports both is essential to keep pace with fast-evolving fraud tactics.
Gupta added that predicting fraudsters’ next moves is a tough challenge, which has led to increased collaboration between banks, regulators, and government agencies. “For instance, banks now form consortia to share information on emerging fraud patterns,” he said. This collaborative approach accelerates fraud detection across institutions and strengthens the broader financial ecosystem.
What’s Next?
According to Kroll’s 2023 Fraud and Financial Crime Report, 70% of financial professionals anticipate a rise in fraud risks in the next 12 months. Thankfully, over two-thirds of financial institutes prioritise technology investments to combat the looming risk of fraud.
There are benefits of using AI fraud detection solutions, like reduced manual review time, better predictions with larger dataset, cost-effectiveness, and more. Yet, these solutions aren’t without challenges. Concerns like false positives, limited human insight, and reduced control over detection processes persist.
To maximise effectiveness, businesses need to choose AI solutions that prioritise accuracy, aiming to keep false positives low. This ensures that fraud detection stays reliable and secure without interrupting legitimate transactions.
The post In the AI Era, Fooling the Banks is Nearly Impossible appeared first on Analytics India Magazine.
